Survival analyses for determining clinical relevance of molecular markers in translational glioblastoma research – a systematic review


Year:

Session type:

Grace Loy1, Beth Fitt1, Edward Christopher2, Paul Brennan2, Michael Poon3
1University of Edinburgh, 2Royal Infirmary of Edinburgh, 3Cancer Research UK (CRUK) Edinburgh Centre

Abstract

Background

Pre-clinical glioblastoma studies can assess the relevance of their findings to patient survival using integrated clinical and genomic data. Validity of univariable analyses requires an assumption that molecular markers are randomly distributed across patient characteristics to mitigate the confounding effects of clinical variables. Multivariable survival analyses adjusting for clinical variables do not change the association if this assumption holds. We aimed to assess this by summarising the types of survival analyses and their results in translational glioblastoma research.  

Method

We systematically searched Medline and Embase Jan 2008 to Feb 2021 for glioblastoma cell line or animal studies validating their molecular markers in The Cancer Genome Atlas (TCGA) or the Chinese Glioma Genome Atlas (CGGA) using survival analyses. Studies that exclusively used genomic data without laboratory findings were excluded. Two reviewers independently assessed study eligibility and extracted data. Data items included patient inclusion criteria, characteristics of survival analyses, and whether molecular markers had a statistically significant association with overall survival. 

Results

Of 1,047 potentially eligible studies, we included 59 pre-clinical glioblastoma studies that tested the association between their molecular markers and survival using TCGA or CGGA data. All studies used TCGA data and 2 also used CGGA data. Sixteen (27%) studies specified their patient inclusion criteria from TCGA for survival analysis. Eight studies exclusively investigated sets of molecular markers, leaving 51 studies reporting 126 molecular markers. Eighteen (31%) studies used multivariable survival analysis in addition to univariable analyses. All molecular markers underwent univariable analyses, of which 12 (10%) molecular markers had additional multivariable survival analyses. In the 13 multivariable analyses on 12 molecular markers, four (31%) markers were associated with survival in the univariable analyses but not in the multivariable analyses.  

Conclusion

Most pre-clinical studies used univariable survival analyses alone in public genomic repositories to assess the relevance of their results to patient survival. Our findings demonstrated that multivariable analyses are needed to account for confounding effects of clinical variables. Using relevant components from reporting guidelines for observational studies can improve the transparency and quality of translational studies. 

Impact statement

Multivariable analyses are needed to account for confounding effects of clinical variables in glioblastoma translational research.